Forming a dataset for fully-supervised learning

Abstract

A computer-implemented method of signal processing comprises providing images. The method comprises for each respective one of at least a subset of the images: applying a weakly-supervised learnt function, the weakly-supervised learnt function outputting respective couples each including a respective localization and one or more respective confidence scores, each confidence score representing a probability of instantiation of a respective object category at the respective localization. The method further comprises determining, based on the output of the weakly-supervised learnt function, one or more respective annotations, each annotation including a respective localization and a respective label representing instantiation a respective object category at the respective localization. The method further comprises forming a dataset including pieces of data, each piece of data including a respective image of the subset and at least a part of the one or more annotations determined for the respective image. This improves the field of object detection.

The invention claimed is: 
     
       1. A computer-implemented method of signal processing comprising:
 obtaining images; 
 for each respective one of at least a subset of the images:
 applying a weakly-supervised learnt function, the weakly-supervised learnt function outputting respective couples each including a respective localization and one or more respective confidence scores, each confidence score representing a probability of instantiation of a respective object category at the respective localization, and 
 determining, based on the output of the weakly-supervised learnt function, one or more respective annotations, each annotation including a respective localization and a respective label representing instantiation a respective object category at the respective localization; and 
 
 forming a dataset including pieces of data, each piece of data including a respective image of the subset and at least a part of the one or more annotations determined for the respective image, 
 wherein the localization of each respective annotation corresponds to one or more localizations outputted by the weakly-supervised learnt function, 
 wherein the object category respective to each respective annotation is an object category having a probability of instantiation, at the one or more localizations outputted by the weakly-supervised learnt function that correspond to the localization of the respective annotation, which is represented by a respective confidence score which is strictly superior to zero, 
 wherein the object category respective to each respective annotation is an object category having a probability of instantiation, at the one or more localizations outputted by the weakly-supervised learnt function that correspond to the localization of the respective annotation, which is represented by a respective confidence score which is superior to a strictly positive threshold, and 
 wherein the threshold has a value which depends on a mean number of objects in the images. 
 
     
     
       2. The method of  claim 1 , wherein the object category respective to each respective annotation is the object category having a probability of instantiation, at the one or more localizations outputted by the weakly-supervised learnt function that correspond to the localization of the respective annotation, which is represented by the highest confidence score. 
     
     
       3. The method of  claim 1 , wherein, for each respective image of at least a part of the subset:
 the respective image is provided with respective initial labels, each initial label representing instantiation of a respective object category in the respective image, and 
 the label of each respective annotation of the respective image representing instantiation of a respective object category corresponding to an initial label of the respective image. 
 
     
     
       4. The method of  claim 1 , wherein the one or more localizations outputted by the weakly-supervised learnt function that correspond to the localization of a respective annotation are identified via a clustering algorithm. 
     
     
       5. The method of  claim 1 , the weakly-supervised learnt function is learnt based on an initial dataset, the initial dataset including initial pieces of data, each initial piece of data including a respective image and a respective annotation, the annotation consisting of a respective set of labels, each label representing instantiation of a respective object category in the respective image. 
     
     
       6. The method of  claim 1 , wherein the method further comprises learning a fully-supervised learnt function based on the formed dataset, the fully-supervised learnt function applying to images and outputting respective couples each including a respective localization and one or more respective confidence scores, each confidence score representing a probability of instantiation of a respective object category at the respective localization. 
     
     
       7. A device comprising:
 a non-transitory storage having stored thereon a data structure, the data structure comprising a computer program including instructions for performing a computer-implemented method of signal processing that when executed by processing circuitry causes the processing circuitry to be configured to: 
 obtain images; 
 for each respective one of at least a subset of the images:
 apply a weakly-supervised learnt function, the weakly-supervised learnt function outputting respective couples each including a respective localization and one or more respective confidence scores, each confidence score representing a probability of instantiation of a respective object category at the respective localization, and 
 determine, based on the output of the weakly-supervised learnt function, one or more respective annotations, each annotation including a respective localization and a respective label representing instantiation a respective object category at the respective localization; and 
 
 form a dataset including pieces of data, each piece of data including a respective image of the subset and at least a part of the one or more annotations determined for the respective image, 
 wherein the localization of each respective annotation corresponds to one or more localizations outputted by the weakly-supervised learnt function, 
 wherein the object category respective to each respective annotation is an object category having a probability of instantiation, at the one or more localizations outputted by the weakly-supervised learnt function that correspond to the localization of the respective annotation, which is represented by a respective confidence score which is strictly superior to zero, 
 wherein the object category respective to each respective annotation is an object category having a probability of instantiation, at the one or more localizations outputted by the weakly-supervised learnt function that correspond to the localization of the respective annotation, which is represented by a respective confidence score which is superior to a strictly positive threshold, and 
 wherein the threshold has a value which depends on a mean number of objects in the images. 
 
     
     
       8. The device of  claim 7 , wherein the non-transitory storage is computer-readable. 
     
     
       9. The device of  claim 7 , wherein the non-transitory storage is a memory, the device further comprising processing circuitry coupled to the memory. 
     
     
       10. A device comprising:
 a non-transitory storage having stored thereon a data structure, the data structure comprising a dataset formed by a computer-implemented method of signal processing that when executed by processing circuitry causes the processing circuitry to be configured to: 
 obtain images; 
 for each respective one of at least a subset of the images:
 apply a weakly-supervised learnt function, the weakly-supervised learnt function outputting respective couples each including a respective localization and one or more respective confidence scores, each confidence score representing a probability of instantiation of a respective object category at the respective localization, and 
 determine, based on the output of the weakly-supervised learnt function, one or more respective annotations, each annotation including a respective localization and a respective label representing instantiation a respective object category at the respective localization; and 
 
 form a dataset including pieces of data, each piece of data including a respective image of the subset and at least a part of the one or more annotations determined for the respective image, 
 wherein the localization of each respective annotation corresponds to one or more localizations outputted by the weakly-supervised learnt function, 
 wherein the object category respective to each respective annotation is an object category having a probability of instantiation, at the one or more localizations outputted by the weakly-supervised learnt function that correspond to the localization of the respective annotation, which is represented by a respective confidence score which is strictly superior to zero, 
 wherein the object category respective to each respective annotation is an object category having a probability of instantiation, at the one or more localizations outputted by the weakly-supervised learnt function that correspond to the localization of the respective annotation, which is represented by a respective confidence score which is superior to a strictly positive threshold, and 
 wherein the threshold has a value which depends on a mean number of objects in the images. 
 
     
     
       11. The device of  claim 10 , wherein the non-transitory storage is computer-readable. 
     
     
       12. The device of  claim 10 , wherein the non-transitory storage is a memory, the device further comprising the processing circuitry. 
     
     
       13. A device comprising:
 a non-transitory storage having stored thereon a data structure, the data structure comprising a fully-supervised learnt function learnable according to a computer-implemented method of signal processing that when executed by processing circuitry causes the processing circuitry to be configured to: 
 obtain images; 
 for each respective one of at least a subset of the images:
 apply a weakly-supervised learnt function, the weakly-supervised learnt function outputting respective couples each including a respective localization and one or more respective confidence scores, each confidence score representing a probability of instantiation of a respective object category at the respective localization, and 
 determine, based on the output of the weakly-supervised learnt function, one or more respective annotations, each annotation including a respective localization and a respective label representing instantiation a respective object category at the respective localization; and 
 
 form a dataset including pieces of data, each piece of data including a respective image of the subset and at least a part of the one or more annotations determined for the respective image, 
 wherein the localization of each respective annotation corresponds to one or more localizations outputted by the weakly-supervised learnt function, 
 wherein the object category respective to each respective annotation is an object category having a probability of instantiation, at the one or more localizations outputted by the weakly-supervised learnt function that correspond to the localization of the respective annotation, which is represented by a respective confidence score which is strictly superior to zero, 
 wherein the object category respective to each respective annotation is an object category having a probability of instantiation, at the one or more localizations outputted by the weakly-supervised learnt function that correspond to the localization of the respective annotation, which is represented by a respective confidence score which is superior to a strictly positive threshold, and 
 wherein the threshold has a value which depends on a mean number of objects in the images. 
 
     
     
       14. The device of  claim 13 , wherein the non-transitory storage is computer-readable. 
     
     
       15. The device of  claim 13 , wherein the non-transitory storage is a memory, the device further comprising the processing circuitry.